Exploring the Cutting Edge: Inside Godel’s AI Lab

Godel Technologies continues to leverage the power of AI as we proudly showcase the Godel AI Lab.

The Godel AI Lab is an innovative place for Godel developers to learn and practice using Large Language Models (LLMs) for tasks. The lab includes four modules, each with real-world examples such as creating an FAQ bot, serving both as an educational platform and a demonstration tool. The lab is currently used internally and provides access to AWS models and encourages experimentation to find cost-effective solutions for clients, pioneering responsible AI practices that ensure the highest quality outcomes.

Katsiaryna Ruksha, Lead Data Scientist, explains more about the AI Lab and how this cutting-edge platform is being utilised at Godel.

Before we dive into the AI Lab, what is the link between data science and artificial intelligence?

I get this question a lot so it would be a good idea to give the definition. Artificial Intelligence is basically any kind of method that attempts to mimic how humans think. This includes robots, models, rule-based algorithms, etc. Machine learning is a part of AI that allows models to learn from data.

The next level is Deep Learning. Deep Learning includes machine learning models with the architecture of deep neural networks. The complexity of architecture makes these models very powerful at solving complicated tasks and working with different data types.

Finally, the next level is generative AI which includes huge transformer-based models that can generate unstructured data.

Data science intersects with AI as data scientists use coding skills, statistics, mathematics and machine learning methods to bring some value to the business.

Can you give an overview of Godel’s AI Lab and its purpose?

The AI Lab is part of the ongoing efforts at Godel to educate and apply AI responsibly. The initial goal was to create resources for developers to build applications using Large Language Models (LLMs) to solve client tasks. At Godel, we are getting more and more requests from our clients to build applications using LLMs. The small data science team at Godel often faces challenges due to limited resources so training AI augmented developers would allow to engage people from other divisions in such projects.

We have a growing AI Community where there are a lot of people trying to use these models. However, we must be cautious about not only writing some framework for the client but also being responsible for the quality. By teaching at AI lab, we want to be sure that when we say that we build something using AI at Godel, then we do it responsibly and with a full understanding of how to address testing and quality assessment.

Being part of the 1st AI Lab, I loved the opportunity to try different kinds of tasks, I enjoyed the structure of the tasks and found the hints section useful.

How do you use the educational and demonstration platform?

The dual purpose of the AI Lab acts as both an educational platform and a demonstration tool and aims to help non-data scientists understand and use AI tools responsibly.

The lab’s structure is split into 4 sections

Extracting information

Create a RAG pipeline

Table Understanding

Finetuning

The four sections correspond to the most frequent tasks that we get from clients and each of the module is solved with a different set of tools.

My goal was to prepare tasks so that developers not only solved them but also spend some effort to analyse the quality of their solutions. This is why each task has two parts: one part is to create a solution, and the second part is always to analyse and answer questions like: what are the metrics that you can use to measure the quality? Are there some specific groups, or classes of data where your model doesn’t perform? How can you address these cases? What can you do to further improve performance?

We also provide access to AWS Bedrock models, so people can either work with open-source models, use Bedrock models or buy access on their own, for example with ChatGPT. We actually encourage people to experiment with many different models, and not to stick to ChatGPT which everyone knows, but to try different stuff and to always think about the way to solve the problem in the cheapest way for the clients. The LLMs are expensive, so for production use they should perform in the cheapest and quickest way. This is why you need to not only stick to the most powerful model but also experiment a lot with alternative models that may get the same quality results at a lower price.

Did you come across any challenges that you’ve overcome?

The biggest challenges were about getting the required resources: obtaining and generating data for the lab that could be relevant to future AI-related projects at Godel and ensuring that we can grant access to Bedrock models for experiments.

Have you received any feedback?

I am pleased to say the first group of the AI Lab has graduated and we gathered a lot of positive feedback. We received an average overall rating of 4.8/5 and 100% of the group said they would recommend AI Labs to others.

What’s next with the AI Lab?

The lab currently focuses on generative AI and Large Language Models. As we get more requests and more projects, we may decide to add new modules. However, currently, the plan is not to continue with Large Language Models but to focus on machine learning itself. Machine learning is much bigger than just LLMs, and there are lots of models that are used for prediction, classification, clustering and other purposes.

The first group of AI Lab has graduated and we are in the middle of mentoring the second group. Resources for the lab are published on SharePoint and available to anyone, but with AI Lab you also have an opportunity to work with a mentor. Currently, Robert Kostrzewski who is another data scientist from the Data division is mentoring the AI lab. We already have a list of people who are waiting for the next groups. If someone is interested in studying at the AI lab, they can contact their talent mentors and get involved in one of the next groups.

The Godel AI Lab is a great way to be part of a new and in-demand area to apply your engineering skills and competencies.